Machine Learning Approach to Automated Early Warning System for Medical Vital Signs Monitoring

Precise and timely detection of deteriorating vital signs is an important aspect of patient safety and clinical intervention. The current standard of monitoring systems lacks automated early warning systems, instead using manual observation to make judgements. This manual approach can lead to delays in detecting critical changes in a patient's condition. We present a novel approach to developing an automated early warning system for vital signs using a hybrid method that combines LSTM (Long Short Term Memory) and XGBoost (Extra Gradient Boost), both methods offer robust predictive modelling that is able to capture the complex and often non-linear relationships inherent in physiological data. This research believes that using a novel technique that combines LSTM and XGBoost advances predictive systems in healthcare-based technology as well as laying the groundwork for even further innovations in early warning systems. The early warning system will evaluate vital signs such as respiratory rate, SpO2 levels, heart rate, body temperature, and pulse which can recognise and predict early signs of clinical deterioration, allowing early intervention and may save a patient’s life. This research will use error metrics such as MAPE, MAE (Mean Absolute Error), MSE, RMSE, and MAD to compare the predicted actual values.